CN111179591A - Road network traffic time sequence characteristic data quality diagnosis and restoration method - Google Patents

Road network traffic time sequence characteristic data quality diagnosis and restoration method Download PDF

Info

Publication number
CN111179591A
CN111179591A CN201911389852.3A CN201911389852A CN111179591A CN 111179591 A CN111179591 A CN 111179591A CN 201911389852 A CN201911389852 A CN 201911389852A CN 111179591 A CN111179591 A CN 111179591A
Authority
CN
China
Prior art keywords
data
time
intersection
road network
intersections
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201911389852.3A
Other languages
Chinese (zh)
Other versions
CN111179591B (en
Inventor
邹开荣
徐甲
谢竞成
丁楚吟
黄贤恒
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yinjiang Technology Co.,Ltd.
Original Assignee
Enjoyor Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Enjoyor Co Ltd filed Critical Enjoyor Co Ltd
Priority to CN201911389852.3A priority Critical patent/CN111179591B/en
Publication of CN111179591A publication Critical patent/CN111179591A/en
Application granted granted Critical
Publication of CN111179591B publication Critical patent/CN111179591B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Tourism & Hospitality (AREA)
  • Strategic Management (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Primary Health Care (AREA)
  • Educational Administration (AREA)
  • Analytical Chemistry (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Traffic Control Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A road network traffic time sequence characteristic data quality diagnosis and restoration method comprises the following steps: s1, obtaining a regional road network topological structure, and constructing a traffic data prediction model; s2, acquiring historical traffic data, training a model and checking the prediction precision of the model; s3, fusing the traffic data acquired by the traffic detector with the traffic data generated, and detecting and repairing the abnormal data in real time; and S4, performing incremental training on the prediction model to ensure the prediction precision of the model. The invention has the beneficial effects that: and predicting a mode of the data on the day and the comprehensive production data of the combined road network data based on historical data and comprehensively judging the quality of the data, and finally realizing high-quality data with reasonable continuous output and strong relevance.

Description

Road network traffic time sequence characteristic data quality diagnosis and restoration method
Technical Field
The invention belongs to the field of traffic information processing, and relates to a road network traffic time sequence characteristic data quality diagnosis and restoration method based on a deep learning method of a graph neural network.
Background
With the development of traffic technology, the use of various algorithms in traffic control becomes more extensive, and the traffic control becomes intelligent and automatic to a great extent.
However, many algorithms can only be applied in a laboratory or in some test point areas, and cannot be fully automatically operated in a large area, which is the fundamental reason that many algorithms such as KNN, CNN, and LSTN have high requirements on data, and unqualified input data can directly cause the unqualified output result of the algorithm. In practical situations, the application of many traffic algorithms is often limited because detectors cannot be guaranteed or the produced data do not meet the actual conditions.
In the existing research, people focus on the quality of data, 1. focus mainly on missing data parts and neglect to check the quality of produced data; 2. for missing portions of data, researchers typically fill in the data using default values for the patching convention or by linear fitting the data over a period of time; 3. for the data of the whole road network, researchers usually only pay attention to the data quality feedback of a certain intersection/road segment/lane, and lack comprehensive quality comparison of adjacent or similar intersection/road segment/lane data. Therefore, it is urgent to construct an effective and robust data production method and a data production link for strictly detecting the data quality.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a road network traffic time sequence characteristic data quality diagnosis and restoration method based on a deep learning method of a neural network, which is a method for predicting the mode of the data of the same day and the comprehensive production data of the combined road network data and comprehensively judging the quality of the data based on historical data, and finally realizes high-quality data with reasonable continuous output and strong relevance.
The technical scheme adopted by the invention is as follows:
a road network traffic time sequence characteristic data quality diagnosis and restoration method comprises the following steps:
s1, obtaining a regional road network topological structure, and constructing a traffic data prediction model;
s2, acquiring historical traffic data, training a model and checking the prediction precision of the model;
s3, fusing the traffic data acquired by the traffic detector with the traffic data generated, and detecting and repairing the abnormal data in real time;
and S4, performing incremental training on the prediction model to ensure the prediction precision of the model.
Further, the traffic data prediction model in step S1 is a convolutional neural network prediction model, and T is presetkAnd constructing a fusion model of the multi-scale prediction result at different time intervals.
Further, an input matrix of the graph convolution neural network prediction model is constructed as follows:
the determined road network system comprises the following steps: g ═ V, E, a, where V denotes the set of all ports in the network G; e represents the whole edge set in the road network G; a represents all intersection adjacency matrixes of the road network G; on the basis of this, so as to
Figure BDA0002344651310000022
The c-th vector on the time slice of the intersection i at the time t is expressed by
Figure BDA0002344651310000023
Representing the set of all vector factors of the intersection i at the moment t; by using
Figure BDA0002344651310000024
To represent the set of all vector factors of all intersections at the time t; by using
Figure BDA0002344651310000021
To represent all historical data of all vector factors for all intersections in the past tau time slices.
Further, the training in step S2 adopts a space-time attention mechanism to learn the temporal relationship between the intersection and the intersection itself, and the specific steps are as follows:
(1) performing double learning on road network data from a space level and a time level by constructing a space-time learning network, and constructing an association relation between road network intersection objects, adjacent time slices and data association between adjacent periodic slices;
(2) performing spatial convolution calculation on each sampling time slice in spatial dimension, and obtaining an incidence relation between intersections by adjusting intersection incidence matrixes so as to determine intersection spatial incidence information; performing convolution calculation on each sampling time slice of each intersection object in a time dimension, and predicting data on a continuous time axis by training a prediction model in a single intersection time dimension;
(3) for three input data vectors with different time granularities, fusion output is carried out on the trained data with different time granularities by giving different weights to the three input data vectors;
(4) and selecting a traditional Mean Square Error (MSE) as a loss function, training the model until the MSE is the minimum, realizing error back propagation by adopting a random gradient descent (SGD) method, and training the model.
Further, the data abnormality in step S3 includes: data missing, suspected noise in the data, and poor data correlation.
Further, for the case of data missing, the specific process of performing the patching processing on the data is as follows:
for data loss of individual intersections, data required by the current time of the intersection object can be predicted through a data model, the intersection data is directly repaired, and the current data of adjacent intersections can be recorded for retraining the next prediction model;
for the condition of data loss of the large-scale intersection, not only data prediction needs to be carried out on the intersection object with the corresponding data loss, but also the current finally repaired data is recorded and is used for retraining a prediction model next time;
if the large-scale data repairing operation is frequently carried out within a certain time, an alarm prompt needs to be carried out on a worker for checking the faults of hardware measures or data input steps.
Further, for the case of data suspected to be noisy, the specific steps required to be recorded are as follows:
1) if only noise data are generated at individual intersections or individual data, the noise data can be directly deleted and repaired by using the data prediction model;
2) if data of suspected noise points frequently appear in a plurality of data production intervals, corresponding records need to be carried out on intersection objects and time produced by the data,
2.1) if the noise data has the characteristics of much noise and no distribution rule, directly identifying the noise data as an abnormal data value, and directly predicting the required time point of the current intersection object and directly restoring the data by using a prediction model;
2.2) if the noisy point data presents the characteristic of a large and regular data distribution, corresponding data comparison needs to be carried out on the change trend of the data attribute of the intersection on the continuous time axis and the topological data association attribute between the intersection object and the peripheral adjacent object; if the adjacent intersections all present the same or similar noise data and the topological correlation attributes exist in the intersection data, manual investigation is carried out on the data source and the data aggregation production process; if for data with the same time granularity, adjacent intersections with similar intersection states have larger difference on noise point data, but for the condition that the data distribution is continuous and data mutation does not exist on the time axis of the corresponding intersection, the data aggregation process and the data source of the detector need to be examined, and the road-gateway connection data are examined to find the problem of poor topological relevance of the intersection data.
Further, for the case of poor data relevance, if the data distribution characteristics of the adjacent intersections have large differences, the data level is used for analyzing the traffic data change trend of the adjacent intersections and the relevance of the traffic data on the time axis, judging whether the data change of the adjacent intersections is normal conduction of the traffic data on the time axis, and if not, inputting by using the latest data and retraining the prediction model to obtain the prediction model based on the current latest data.
Further, whether the data change of the adjacent intersection is normal transmission of the traffic data on the time axis can be judged by the congestion alarm amount, which is specifically as follows:
selecting M adjacent intersections around the single intersection, and determining the actual time delay △ T of the two objects by comparing the time difference △ T when the continuous alarm times between the single intersection and the adjacent intersections reach the preset threshold value with the maximum continuous alarm times*comparing the time difference with the time difference △ T' of the data change trend between two intersections predicted by the prediction model,
Figure BDA0002344651310000041
Figure BDA0002344651310000051
△T*＝△T*S
Figure BDA0002344651310000052
wherein the content of the first and second substances,
Figure BDA0002344651310000053
the alarm starting time for the intersection a is set,
Figure BDA0002344651310000054
alarming starting time for the intersection b, and selecting the number of time intervals for congestion alarming data by n; t isa,iFor crossing a alarm duration, Tb,iAlarm holder for b intersectionThe duration of the time period is longer,
Figure BDA0002344651310000055
the maximum continuous alarm frequency of the intersection a,
Figure BDA0002344651310000056
the maximum continuous alarm frequency of the intersection b is obtained;
Figure BDA0002344651310000057
represents the corresponding data of the b-th intersection in the T + f-th group prediction data on the time axis,
Figure BDA0002344651310000058
for predicting data corresponding to the intersection a in the T-th group on the time axis, TτFor data update time granularity, Threshold is a set Threshold for predicting data difference,
if △ T*if the difference in value from Δ T' is less than the corresponding threshold, it can be assumed that the data characteristic between intersections changes to normal transmission of data on the time axis.
Further, whether the data change of the adjacent intersection is normal transmission of the traffic data on the time axis can be judged by the speed of the adjacent intersection, which is specifically as follows:
the intersection with more advanced data change characteristics is used as a current data template, a time period with larger speed change trend every day is set as an experimental time interval, corresponding time speed values of the intersection with slower data change characteristics are searched for by accumulating corresponding delta T 'on a data time axis of the intersection with more backward data change characteristics, △ T' is a time difference of data change trends between two intersections predicted by a prediction model, and if the frequency that the difference value of the speed values of the two intersections in the selected experimental time interval exceeds a speed fluctuation interval (namely the threshold range of positive and negative change of data) is smaller than a set threshold, the change of the data characteristics between the intersections can be determined as normal conduction influence time of the data on the time axis.
Further, step S3 includes, after the data is repaired by using the prediction model, checking the quality of the data produced on the day, passing the qualified data, and feeding back the unqualified data by using a method of reproducing or retraining the data prediction model and checking again, wherein the data quality checking method includes:
Figure BDA0002344651310000061
Figure BDA0002344651310000062
ξ is the sampling frequency of certain traffic data at the intersection every day;
Figure BDA0002344651310000063
predicting the predicted value of the data for the data prediction model at the ith sampling;
Figure BDA0002344651310000064
producing a real-time value of the data generated for the data at the ith sampling; threshold is the Threshold range within which the two differ by an acceptable amount; m is the deviation between the data produced under the intersection data sampling frequency and each sampling frequency and the data prediction model prediction data, and the smaller the numerical value is, the better the data quality is considered; otherwise, the data quality is considered to be poor.
Further, intersection data with obviously poor data quality is further analyzed, if the periodic poor data quality condition occurs in a certain intersection or a certain sub-area, the actual condition of the intersection or the area is firstly analyzed, if the intersection or the area has the same periodic occurrence activity in a real road network, a new data prediction model is constructed by using the latest produced real-time data, and the data of the day is predicted, repaired and detected by using different data prediction models according to the actual condition of the day;
if the intersection is in a special condition after the actual road network condition is observed, the deviation between the data and the prediction model is large in the production time and on the continuous time axis, and similar conditions occur in a certain number of intersections in a corresponding area by integrating the data change conditions of the intersection and the surrounding intersections, a method for retraining the data prediction model and monitoring the quality of the intersection with the real-time production data of the day again need to be considered, and an actual and objective data quality detection result is generated.
The invention has the beneficial effects that: and predicting a mode of the data on the day and the comprehensive production data of the combined road network data based on historical data and comprehensively judging the quality of the data, and finally realizing high-quality data with reasonable continuous output and strong relevance.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
FIG. 2 is a schematic diagram of different vector factors at an intersection of the present invention.
FIG. 3 is a schematic structural diagram of a graph convolution neural network training process of the present invention.
FIG. 4 is a schematic structural diagram of a neural network training model of the present invention.
FIG. 5 is a schematic diagram of the irregular noise distribution of the present invention.
FIG. 6 is a schematic diagram of the regular noise distribution of the present invention.
FIG. 7 is a comparison diagram of the speed relevance of the adjacent crossing of the present invention.
FIG. 8 is a schematic diagram of real-time data and data model trend in accordance with the present invention.
Detailed Description
The present invention is further illustrated by the following examples, which are not intended to limit the invention to these embodiments. It will be appreciated by those skilled in the art that the present invention encompasses all alternatives, modifications and equivalents as may be included within the scope of the claims.
Referring to fig. 1, the invention provides a road network traffic time series characteristic data quality diagnosis and repair method, which comprises the following steps:
s1, obtaining a regional road network topological structure, and constructing a traffic data prediction model;
s2, acquiring historical traffic data, training a model and checking the prediction precision of the model;
s3, fusing the traffic data acquired by the traffic detector with the traffic data generated, and detecting and repairing the abnormal data in real time;
and S4, performing incremental training on the prediction model to ensure the prediction precision of the model.
The specific process is as follows:
in an actual road network system, under the condition that no significant or unpredictable things happen, 1. dynamic data of each intersection has continuous and predictable attributes on a continuous time axis based on the dynamic data; 2. the dynamic data of each intersection has similar and predictable attributes at corresponding time points based on the continuous period of the dynamic data; 3. the dynamic data of each intersection object of the whole road network has road network topological relevance and predictable attributes based on other objects. Therefore, for a single intersection data, not only the correlation of the intersection data on the time axis is needed to be related, but also the correlation relationship between intersections as a part of the whole road network is considered, so that the current intersection data is correctly and reasonably predicted by integrating time and geographic information.
For each intersection, in unit time, several required indicators, such as speed, flow, saturation, etc., need to be detected as measurement factors. In an actual road network, since the measurement factor of a certain intersection also changes continuously on a continuous time axis, as shown in fig. 2, a continuously time-varying vector can be constructed for each intersection by recording the plurality of measurement factors.
For a certain road network system G ═ (V, E, a), where V denotes the set of all the road ports in the road network G; e represents the whole edge set in the road network G; a denotes the entire intersection adjacency matrix of the road network G. On the basis of this, so as to
Figure BDA0002344651310000081
The c-th vector on the time slice of the intersection i at the time t can be further used
Figure BDA0002344651310000082
Representing the set of all vector factors of the intersection i at the moment t; further, it can be used
Figure BDA0002344651310000083
To represent the set of all vector factors of all intersections at the time t; further, it can be used
Figure BDA0002344651310000084
To represent all historical data of all vector factors for all intersections in the past tau time slices.
The traffic data are sampled at different time intervals, traffic laws of different scales can be observed, and in order to improve the accuracy of a prediction result, the invention provides a multi-scale prediction result fusion model.
Sampling the original frequency data by TkResampling the interval scale to obtain a training data set under the scale
Figure BDA0002344651310000085
For example:
I. sampling at 10 min intervals
Figure BDA0002344651310000086
The data of the next time slice can be predicted;
II. Sampling at 1 hour intervals
Figure BDA0002344651310000091
The data at this time point for the next hour can be predicted;
III, sampling at 1 day intervals
Figure BDA0002344651310000092
The data at this time point in the tomorrow may be predicted.
However, for the road network data, not only the intersection data is predicted in time, but also the data relation caused by the topological relation between intersections occupies a more important component on a single time slice, and the interaction between intersections plays a very important role in the interaction of dynamic data in the spatial dimension. The invention adopts a space-time attention mechanism to learn the time relation between intersections and the intersections, and the intersections per se, and the method comprises the following specific steps:
(1) the spatial attention mechanism is as follows:
Figure BDA0002344651310000093
Figure BDA0002344651310000094
wherein
Figure BDA0002344651310000095
Represents TkInputting data of an r sampling period under the sampling frequency; cr-1An r-th data channel of the data channels representing the input data; vs，bs∈RN×N,
Figure BDA0002344651310000096
Representing learning parameters in a spatial attention mechanism; sigma is an activation parameter; si,j,S′i,jAnd respectively representing the incidence relation between the intersection i and the intersection j in the road network matrix before and after the convolutional network learning.
(2) Time attention mechanism:
Figure BDA0002344651310000097
Figure BDA0002344651310000098
similar to the spatial attention mechanism described above,
Figure BDA0002344651310000099
U1∈RN,
Figure BDA00023446513100000910
representing a learning parameter in a temporal attention mechanism; sigma is an activation parameter; ei,j,E′i,jAnd respectively representing the association relationship between the time i and the time j before and after the convolutional network learning.
Therefore, as shown in fig. 2, the spatial-temporal learning network is constructed to perform dual learning on the road network data from the spatial level and the temporal level, so as to construct the association relationship between road network intersection objects and the data association between adjacent time slices and adjacent period slices.
(3) Convolution of spatial dimensions
Based on the idea of atlas theory, every point on the map can be regarded as an input of a signal. Therefore, in each sampling time slice, the data attributes on the graph need to be converted into an algebraic form, and then the data attributes of the graph need to be analyzed and judged. Inspired by the theory of maps, the attributes of each map can be represented by a laplacian matrix and its eigenvalues, as shown in equations (5) and (6):
L＝D-A (5)
L＝IN-D-1/2AD-1/2(6)
wherein L is a Laplace matrix of a graph corresponding to the road network; d is a degree matrix of the road network diagram; a is an adjacent matrix of the road network graph; i isNIs an identity matrix.
With knowledge of linear algebra, we can decompose the laplacian matrix into the following form:
L＝UΛUT(7)
wherein U is a Fourier transform matrix, and Λ is a characteristic value of an L matrix;
and all intersection data input x on the graph at a certain time ttThe fourier transform of (d) can be expressed as:
Figure BDA0002344651310000101
Figure BDA0002344651310000102
thus, x on the graph GtThe data input of (a) can be filtered by the convolution operator as:
gθ*Gxt＝gθ(L)xt＝gθ(UΛUT)xt＝Ugθ(Λ)UTxt(10)
equation 10 can be understood as calculating by directly converting the signal input data on the graph into the frequency domain, where G represents the convolution operation and G represents the frequency domainθIs a convolution kernel. Therefore, in terms of node relevance, it can be understood that the data of each intersection is updated by the data of k-1 intersections in the road network.
(4) Convolution of time dimension
Similar to the above-described spatial dimension graph convolution, the graph convolution of the temporal dimension can also be understood similarly:
Figure BDA0002344651310000111
the data at the time t on the graph is updated from the data at the previous time t-1, such as G in formula (11) representing convolution operation, phi is a time convolution kernel parameter, GθFor convolution kernel, ReLU is a rectifying linear unit to activate the module;
and finally, carrying out final integrated output. For data inputs of different time dimensions (e.g., weekly period, daily period, adjacent period), the weight of each dimension data is completely different.
Figure BDA0002344651310000112
As shown in the above-mentioned (12),
Figure BDA0002344651310000113
learning parameters (weights) representing the three different dimensional time data inputs,
Figure BDA0002344651310000114
representing data input in three different dimensions, Θ is the hadamard product.
As shown in fig. 3, in the training process, in the spatial dimension, spatial convolution calculation is performed on each sampling time slice, and the incidence relation between intersections is obtained by adjusting the intersection incidence matrix, so as to determine intersection spatial incidence information; and in the time dimension, performing convolution calculation on each sampling time slice of each intersection object, and predicting data on a continuous time axis by training a prediction model in the time dimension of a single intersection.
(5) Data fusion layer
And for three input data vectors with different time granularities, performing fusion output on the trained data with different time granularities by giving different weights to the three input data vectors.
(6) Computation of loss function
The loss function selects the traditional Mean Square Error (MSE), and the training goal of the model is to minimize the MSE:
Figure BDA0002344651310000115
where N is the sample size of the training,
Figure BDA0002344651310000121
is the ith group of data in the prediction result,
Figure BDA0002344651310000122
is the ith set of data in the actual result. And a random gradient descent (SGD) method is adopted to realize error back propagation, and the model is trained. Compared with the traditional batch gradient descent method, the random gradient descent method has a faster updating speed.
The graph convolution neural network training model of the present invention is shown in FIG. 4.
After the neural network prediction model is completed, the method can be applied to the production process and the self-checking process of data. For real-time data which is produced on line and is based on a road network detector, each new data is not only an expression form of data continuity existing between the data with the previous time granularity and the same intersection object in theory, but also a topological data relevance existing in spatial relevance with surrounding intersection objects.
In a real-time data production environment, the following problems often exist: 1. individual intersection object data is missing; 2. large-scale intersection object data is missing; 3. data suspected noise points; 4. poor data relevance and the like.
First, in the case of data loss, a repair process is required for the data.
For data loss of individual intersections, data required by the current time of the intersection object can be predicted through a data model, the intersection data is directly repaired, and the current data of adjacent intersections can be recorded for retraining the next prediction model;
for the condition of data loss of the large-scale intersection, not only data prediction needs to be carried out on the intersection object with the corresponding data loss, but also the current finally repaired data is recorded and is used for retraining a prediction model next time; if the large-scale data repairing operation is frequently carried out within a certain time, an alarm prompt needs to be carried out on a worker for checking the faults of hardware measures or data input steps.
For the case of data suspected to be noisy, it needs to be recorded.
1) If only noise data are generated at individual intersections or individual data, the noise data can be directly deleted and repaired by using the data prediction model;
2) if data of suspected noise points frequently appear in a plurality of next data production intervals, the corresponding record needs to be carried out on the intersection objects and the time of the data production,
2.1) if the noisy point data has the characteristics of much and no distribution rule, as shown in fig. 5, the noisy point data can be directly identified as an abnormal data value, and the data prediction model directly performs data prediction on the demand time point of the current intersection object and directly performs data restoration.
2.2) if the noisy point data presents the characteristic of a large and regular distribution of data quantity, as shown in fig. 6, it is necessary to perform corresponding data comparison on the change trend of the data attribute on the time axis of the intersection itself and the topological data association attribute between the intersection object and the surrounding neighboring objects,
if the adjacent intersections present the same or similar noise data and the topological correlation attributes exist based on the intersection data, the data sources and the data aggregation production process are manually checked according to the existing related steps; if for data with the same time granularity, adjacent intersections with similar intersection states have larger difference on noise point data, but for the condition that the data distribution is continuous and data mutation does not exist on the time axis of the corresponding intersection, the data aggregation process and the data source of the detector need to be examined, and the road-gateway connection data are examined to find the problem of poor topological relevance of the intersection data.
for the problem of poor data relevance, if the data distribution characteristics of adjacent intersections have large differences, as shown in fig. 7, in a normal situation, the same or similar intersection attribute factor variation trends should exist between adjacent intersections, if the intersection data variation trends have large differences, the intersection variation trends should be analyzed from a data layer, and if the data variation trends between intersections meet the requirement that the variation trends are different from each other on a time axis within a certain time interval, the relevance of each piece of traffic data on the time axis needs to be calculated.
The embodiment takes speed and congestion warning amount as examples for calculation and explanation:
selecting M adjacent intersections around the single intersection, and determining the actual time delay △ T of the two objects by comparing the time difference △ T when the continuous alarm times between the single intersection and the adjacent intersections reach the preset threshold value with the maximum continuous alarm times*The time value is used as a delay factor.
Alarm time difference, alarm similarity calculation and time delay are shown in (14), (15) and (16):
Figure BDA0002344651310000141
Figure BDA0002344651310000149
△T*＝△T*S (16)
Figure BDA0002344651310000142
wherein in (14)
Figure BDA0002344651310000143
The alarm starting time for the intersection a is set,
Figure BDA0002344651310000144
the method comprises the steps that (1) alarm starting time is set for an intersection b, n is a time period selected for congestion alarm data, 7 practical sections with high alarm frequency and large alarm frequency change amplitude are used as alarm data extraction time intervals for 7 hours, 8 hours, 9 hours, 16 hours, 17 hours, 18 hours and 19 hours in the text respectively, and n is 7; (15) middle Ta,iFor crossing a alarm duration, Tb,iThe alarm duration time of the intersection b is set,
Figure BDA0002344651310000145
the maximum continuous alarm frequency of the intersection a,
Figure BDA0002344651310000146
the threshold value of the continuous alarming times is set to be 3, △ T' is the time difference of data change trend between two intersections predicted by the prediction model,
Figure BDA0002344651310000147
represents the corresponding data of the b-th intersection in the T + f-th group prediction data on the time axis,
Figure BDA0002344651310000148
for predicting data corresponding to the intersection a in the T-th group on the time axis, TτThreshold is a set Threshold for the predicted data difference for the data update time granularity.
if △ T*if the numerical difference from the value of △ T' is less than the corresponding threshold value, the data characteristic change between the intersections can be determined as the normal transmission time of the data on the time axis, if the numerical difference from*if the numerical difference between the data prediction model and the value delta T' is larger, the latest data is adopted for inputting the data prediction model, and the prediction model is retrained, so that the data prediction model based on the current latest data is obtained.
if the speed is taken as data input, whether data mutation occurs or the data change difference of adjacent intersections is overlarge is judged according to the data change between the adjacent intersections, the intersection with the data change characteristic being more advanced is taken as a current data template, a time period with a larger speed change trend every day is set as an experimental time interval (for example, the time interval from 6 to 8 days every day and from 16 to 17 days is generally a time intersection point from a flat peak to a peak), corresponding time speed values of the intersection with the slow data change characteristic are searched by accumulating corresponding △ T' on a time axis of intersection data with the data change characteristic being more backward, if the frequency of the difference value of the two speed values in the selected experimental time interval exceeding a speed fluctuation interval (namely the threshold range of positive and negative change of the data) is smaller than a set threshold, the data characteristic change between the intersections can be determined as the normal conduction influence time of the data on the time axis, otherwise, a new data prediction model is trained by adopting a new data prediction model for corresponding data repairing model.
And obtaining a new data prediction model after data repairing, checking the quality of data produced on the same day, passing qualified data, and feeding back unqualified data by adopting a mode of reproducing or retraining the data prediction model and checking again.
After the intersections affected by unpredictable events (including emergency events such as road section maintenance, traffic accidents, special case tasks and the like) are eliminated, the data of the intersections which normally run for one day are analyzed. In general, the data produced and the data template should be continuous lines which normally fluctuate mutually in the same time period, and the values of the two are basically the same or the difference value is a certain threshold value, and the data change is shown in fig. 8.
The intersection data characteristic change trend takes the change trend of the data predicted by the data prediction model as a template, and fluctuation change is carried out around the template, so that the data quality of the production data can be correspondingly judged by using the following method:
Figure BDA0002344651310000151
Figure BDA0002344651310000152
ξ is the sampling frequency of certain traffic data at the intersection every day;
Figure BDA0002344651310000153
predicting the predicted value of the data for the data prediction model at the ith sampling;
Figure BDA0002344651310000154
producing a real-time value of the data generated for the data at the ith sampling; threshold is the Threshold range within which the two differ by an acceptable amount; m is the deviation between the data produced under the intersection data sampling frequency and each sampling frequency and the data prediction model prediction data, and the smaller the numerical value is, the better the data quality is considered; otherwise, the data quality is considered to be poor. The quantization processing based on the sampling frequency is performed on the quality of the data by the formulas (18) and (19), and thus, the quality of the data is not only judged in the past with a fuzzy word eye, such as good quality, generally poor quality, and the like.
Further analyzing intersection data with obviously poor data quality, if the periodic poor data quality condition occurs in a certain intersection or a certain sub-area, firstly analyzing the actual condition of the intersection or the area, if the intersection or the area has the same periodic occurrence activities (such as market opening, road maintenance, entertainment activities and the like) in a real road network, constructing a new data prediction model by using the latest produced real-time data, and predicting, repairing and detecting the data of the same day by using different data prediction models according to the actual condition of the same day;
if special conditions, such as traffic accidents and special guarantees, occur at intersections after the actual road network condition is observed, the deviation between data and the prediction model is large in the production time and on the continuous time axis, and similar conditions occur at a certain number of intersections in a corresponding area by integrating the data change conditions of the intersections and the surrounding intersections, a method for retraining the data prediction model and monitoring the quality of the data with the real-time production data of the day again need to be considered, and an actual and objective data quality detection result is generated.
According to the method, the mode of the data on the day and the comprehensive production data of the combined road network data is predicted based on the historical data, the quality of the data is comprehensively judged, and finally high-quality data with reasonable continuous output and strong relevance is realized.

Claims (12)

1. A road network traffic time sequence characteristic data quality diagnosis and restoration method comprises the following steps:
s1, obtaining a regional road network topological structure, and constructing a traffic data prediction model;
s2, acquiring historical traffic data, training a model and checking the prediction precision of the model;
s3, fusing the traffic data acquired by the traffic detector with the traffic data generated, and detecting and repairing the abnormal data in real time;
and S4, performing incremental training on the prediction model to ensure the prediction precision of the model.
2. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 1, characterized in that: the traffic data prediction model in step S1 is a convolutional neural network prediction model, and T is presetkAnd constructing a fusion model of the multi-scale prediction result at different time intervals.
3. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 2, characterized in that: the input matrix of the graph convolution neural network prediction model is constructed as follows:
the determined road network system comprises the following steps: g ═ V, E, a, where V denotes the set of all ports in the network G; e represents the whole edge set in the road network G; a represents all intersection adjacency matrixes of the road network G; on the basis of this, so as to
Figure FDA0002344651300000011
The c-th vector on the time slice of the intersection i at the time t is expressed by
Figure FDA0002344651300000012
Representing the set of all vector factors of the intersection i at the moment t; by using
Figure FDA0002344651300000013
To represent the set of all vector factors of all intersections at the time t; by using
Figure FDA0002344651300000014
To represent all historical data of all vector factors for all intersections in the past tau time slices.
4. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 2, characterized in that: the training in step S2 adopts a space-time attention mechanism to learn the temporal relationship between intersections and the intersection itself, and the specific steps are as follows:
(1) performing double learning on road network data from a space level and a time level by constructing a space-time learning network, and constructing an association relation between road network intersection objects, adjacent time slices and data association between adjacent periodic slices;
(2) performing spatial convolution calculation on each sampling time slice in spatial dimension, and obtaining an incidence relation between intersections by adjusting intersection incidence matrixes so as to determine intersection spatial incidence information; performing convolution calculation on each sampling time slice of each intersection object in a time dimension, and predicting data on a continuous time axis by training a prediction model in a single intersection time dimension;
(3) for input data vectors of different time intervals, performing fusion output on the trained data of different time granularities by giving different weights to the input data vectors;
(4) and selecting a mean square error by the loss function, training the model until the mean square error MSE is minimum, realizing error back propagation by adopting a random gradient descent method, and training the model.
5. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 1, characterized in that: the data abnormality in step S3 includes: data missing, suspected noise in the data, and poor data correlation.
6. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 5, characterized in that: for the case of data missing, the specific process of performing the patching processing on the data is as follows:
for data loss of individual intersections, the data required by the current time of the intersection object can be predicted through the prediction model, the intersection data is directly repaired, and the current data of the adjacent intersections can be recorded for retraining the prediction model at the next time;
for the condition of data loss of the large-scale intersection, not only data prediction needs to be carried out on the intersection object with the corresponding data loss, but also the current finally repaired data is recorded and is used for retraining a prediction model next time; if the large-scale data repairing operation is frequently carried out within a certain time, an alarm prompt needs to be carried out on a worker for checking the faults of hardware measures or data input steps.
7. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 5, characterized in that: for the suspected noise of the data, the specific steps required to be recorded are as follows:
1) if only noise data are generated at individual intersections or individual data, the noise data can be directly deleted and repaired by using the data prediction model;
2) if data of suspected noise points frequently appear in a plurality of data production intervals, corresponding records need to be carried out on intersection objects and time produced by the data,
2.1) if the noise data has the characteristics of much noise and no distribution rule, directly identifying the noise data as an abnormal data value, and directly predicting the required time point of the current intersection object and directly restoring the data by using a prediction model;
2.2) if the noisy point data presents the characteristic of a large and regular data distribution, corresponding data comparison needs to be carried out on the change trend of the data attribute of the intersection on the continuous time axis and the topological data association attribute between the intersection object and the peripheral adjacent object; if the adjacent intersections all present the same or similar noise data and the topological correlation attributes exist in the intersection data, manual investigation is carried out on the data source and the data aggregation production process; if for data with the same time granularity, adjacent intersections with similar intersection states have larger difference on noise point data, but for the condition that the data distribution is continuous and data mutation does not exist on the time axis of the corresponding intersection, the data aggregation process and the data source of the detector need to be examined, and the road-gateway connection data are examined to find the problem of poor topological relevance of the intersection data.
8. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 5, characterized in that: for the condition of poor data relevance, if the data distribution characteristics of the adjacent intersections have large differences, the change trend of the traffic data of the adjacent intersections and the relevance of the traffic data on the time axis are analyzed from the data level, whether the data change of the adjacent intersections is normal conduction of the traffic data on the time axis is judged, and if not, the latest data is adopted for inputting and retraining the prediction model, so that the prediction model based on the current latest data is obtained.
9. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 8, characterized in that: whether the data change of the adjacent intersection is normal transmission of traffic data on a time axis can be judged through the congestion alarm amount, and the method specifically comprises the following steps:
m adjacent intersections around the single intersection are selected, and the actual time delay △ T of the two objects is determined through the time difference △ T and the alarm similarity S when the continuous alarm times between the single intersection and the adjacent intersections reach the preset threshold value*comparing the time difference with the time difference △ T' of the data change trend between two intersections predicted by the prediction model,
Figure FDA0002344651300000041
Figure FDA0002344651300000042
△T*＝△T*S
Figure FDA0002344651300000043
wherein the content of the first and second substances,
Figure FDA0002344651300000044
the alarm starting time for the intersection a is set,
Figure FDA0002344651300000045
alarming starting time for the intersection b, and selecting the number of time intervals for congestion alarming data by n; t isa,iFor crossing a alarm duration, Tb,iThe alarm duration time of the intersection b is set,
Figure FDA0002344651300000046
the maximum continuous alarm frequency of the intersection a,
Figure FDA0002344651300000047
the maximum continuous alarm frequency of the intersection b is obtained;
Figure FDA0002344651300000048
represents the corresponding data of the b-th intersection in the T + f-th group prediction data on the time axis,
Figure FDA0002344651300000049
for predicting data corresponding to the intersection a in the T-th group on the time axis, TτUpdating time granularity for data, wherein Threshold is a set Threshold value of a predicted data difference value;
if △ T*if the difference in value from Δ T' is less than the corresponding threshold, it can be assumed that the data characteristic between intersections changes to normal transmission of data on the time axis.
10. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 8, characterized in that: whether the data change of the adjacent intersection is normal conduction of traffic data on a time axis can be judged by the speed of the adjacent intersection, and the specific steps are as follows:
the intersection with more advanced data change characteristics is used as a current data template, a time period with larger speed change trend every day is set as an experimental time interval, corresponding time speed values of the intersection with slower data change characteristics are searched for by accumulating corresponding delta T 'on a data time axis of the intersection with more backward data change characteristics, △ T' is a time difference of data change trends between two intersections predicted by a prediction model, and if the number of times that the difference value of the speed values of the two intersections exceeds a speed fluctuation interval in the selected experimental time interval is smaller than a set threshold value, the data characteristic change between the intersections can be determined as the normal conduction influence time of the data on the time axis.
11. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 1, characterized in that: step S3 further includes, after data patch is performed by using the prediction model, checking the quality of data produced on the day, passing qualified data, and feeding back unqualified data by using a method of reproducing or retraining the data prediction model and checking again, where the data quality checking method includes:
Figure FDA0002344651300000051
Figure FDA0002344651300000052
ξ is the sampling frequency of certain traffic data at the intersection every day;
Figure FDA0002344651300000053
predicting the predicted value of the data for the data prediction model at the ith sampling;
Figure FDA0002344651300000054
producing a real-time value of the data generated for the data at the ith sampling; threshold is the Threshold range within which the two differ by an acceptable amount; m is the deviation between the data produced under the intersection data sampling frequency and each sampling frequency and the data prediction model prediction data, and the smaller the numerical value is, the better the data quality is considered; otherwise, the data quality is considered to be poor.
12. The road network traffic time series characteristic data quality diagnosis and restoration method according to claim 11, wherein: further analyzing intersection data with obviously poor data quality, if the periodic poor data quality condition occurs in a certain intersection or a certain sub-area, firstly analyzing the actual condition of the intersection or the area, if the intersection or the area has the same periodic occurrence in a real road network, constructing a new data prediction model by using the latest produced real-time data, and predicting, repairing and detecting the data of the day by using different data prediction models according to the actual condition of the day;
if the intersection is in a special condition after the actual road network condition is observed, the deviation between the data and the prediction model is large in the production time and on the continuous time axis, and similar conditions occur in a certain number of intersections in a corresponding area by integrating the data change conditions of the intersection and the surrounding intersections, a method for retraining the data prediction model and monitoring the quality of the intersection with the real-time production data of the day again need to be considered, and an actual and objective data quality detection result is generated.
CN201911389852.3A 2019-12-30 2019-12-30 Road network traffic time sequence characteristic data quality diagnosis and restoration method Active CN111179591B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911389852.3A CN111179591B (en) 2019-12-30 2019-12-30 Road network traffic time sequence characteristic data quality diagnosis and restoration method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911389852.3A CN111179591B (en) 2019-12-30 2019-12-30 Road network traffic time sequence characteristic data quality diagnosis and restoration method

Publications (2)

Publication Number Publication Date
CN111179591A true CN111179591A (en) 2020-05-19
CN111179591B CN111179591B (en) 2021-02-26

Family

ID=70655790

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911389852.3A Active CN111179591B (en) 2019-12-30 2019-12-30 Road network traffic time sequence characteristic data quality diagnosis and restoration method

Country Status (1)

Country Link
CN (1) CN111179591B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111785014A (en) * 2020-05-26 2020-10-16 浙江工业大学 Road network traffic data restoration method based on DTW-RGCN
CN112309112A (en) * 2020-09-10 2021-02-02 浙江工业大学 Traffic network data restoration method based on GraphSAGE-GAN
CN112380206A (en) * 2020-11-25 2021-02-19 浙江浙大中控信息技术有限公司 Diagnosis and repair method of traffic time sequence data
CN112910695A (en) * 2021-01-22 2021-06-04 湖北工业大学 Network fault prediction method based on global attention time domain convolutional network
CN114091357A (en) * 2022-01-19 2022-02-25 杭州海康威视数字技术股份有限公司 Traffic flow prediction method, traffic flow prediction device, electronic device, and storage medium
CN114356998A (en) * 2022-01-08 2022-04-15 浙江力石科技股份有限公司 Big data tourism resource data compensation method

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001022722A (en) * 1999-07-05 2001-01-26 Nippon Telegr & Teleph Corp <Ntt> Method and device for finding number law conditioned by qualitative variable and storage medium stored with finding program for number law conditioned by qualitative variable
CN101604478A (en) * 2009-06-18 2009-12-16 北京九州联宇信息技术有限公司 A kind of disposal route of dynamic information and system
CN103971520A (en) * 2014-04-17 2014-08-06 浙江大学 Traffic flow data recovery method based on space-time correlation
CN105764162A (en) * 2016-05-10 2016-07-13 江苏大学 Wireless sensor network abnormal event detecting method based on multi-attribute correlation
CN106971547A (en) * 2017-05-18 2017-07-21 福州大学 A kind of Short-time Traffic Flow Forecasting Methods for considering temporal correlation
CN108898829A (en) * 2018-06-07 2018-11-27 重庆邮电大学 The dynamic short-time traffic flow forecast system with Sparse is divided for the indifference opposite sex
CN109544926A (en) * 2018-12-06 2019-03-29 中国公路工程咨询集团有限公司 A kind of magnitude of traffic flow restorative procedure based on intersections correlation
JP2019100136A (en) * 2017-12-06 2019-06-24 株式会社東芝 Road maintenance management system, road maintenance management method and computer program
CN110083593A (en) * 2019-04-23 2019-08-02 中国大唐集团科学技术研究院有限公司华东电力试验研究院 The cleaning of power station operating parameter and restorative procedure, repair system
CN110164128A (en) * 2019-04-23 2019-08-23 银江股份有限公司 A kind of City-level intelligent transportation analogue system
CN110533932A (en) * 2019-08-06 2019-12-03 银江股份有限公司 A kind of City-level universe traffic signals recommended method and system based on rolling optimization
CN110570651A (en) * 2019-07-15 2019-12-13 浙江工业大学 Road network traffic situation prediction method and system based on deep learning

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001022722A (en) * 1999-07-05 2001-01-26 Nippon Telegr & Teleph Corp <Ntt> Method and device for finding number law conditioned by qualitative variable and storage medium stored with finding program for number law conditioned by qualitative variable
CN101604478A (en) * 2009-06-18 2009-12-16 北京九州联宇信息技术有限公司 A kind of disposal route of dynamic information and system
CN103971520A (en) * 2014-04-17 2014-08-06 浙江大学 Traffic flow data recovery method based on space-time correlation
CN105764162A (en) * 2016-05-10 2016-07-13 江苏大学 Wireless sensor network abnormal event detecting method based on multi-attribute correlation
CN106971547A (en) * 2017-05-18 2017-07-21 福州大学 A kind of Short-time Traffic Flow Forecasting Methods for considering temporal correlation
JP2019100136A (en) * 2017-12-06 2019-06-24 株式会社東芝 Road maintenance management system, road maintenance management method and computer program
CN108898829A (en) * 2018-06-07 2018-11-27 重庆邮电大学 The dynamic short-time traffic flow forecast system with Sparse is divided for the indifference opposite sex
CN109544926A (en) * 2018-12-06 2019-03-29 中国公路工程咨询集团有限公司 A kind of magnitude of traffic flow restorative procedure based on intersections correlation
CN110083593A (en) * 2019-04-23 2019-08-02 中国大唐集团科学技术研究院有限公司华东电力试验研究院 The cleaning of power station operating parameter and restorative procedure, repair system
CN110164128A (en) * 2019-04-23 2019-08-23 银江股份有限公司 A kind of City-level intelligent transportation analogue system
CN110570651A (en) * 2019-07-15 2019-12-13 浙江工业大学 Road network traffic situation prediction method and system based on deep learning
CN110533932A (en) * 2019-08-06 2019-12-03 银江股份有限公司 A kind of City-level universe traffic signals recommended method and system based on rolling optimization

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111785014A (en) * 2020-05-26 2020-10-16 浙江工业大学 Road network traffic data restoration method based on DTW-RGCN
CN112309112A (en) * 2020-09-10 2021-02-02 浙江工业大学 Traffic network data restoration method based on GraphSAGE-GAN
CN112380206A (en) * 2020-11-25 2021-02-19 浙江浙大中控信息技术有限公司 Diagnosis and repair method of traffic time sequence data
CN112910695A (en) * 2021-01-22 2021-06-04 湖北工业大学 Network fault prediction method based on global attention time domain convolutional network
CN112910695B (en) * 2021-01-22 2022-07-19 湖北工业大学 Network fault prediction method based on global attention time domain convolution network
CN114356998A (en) * 2022-01-08 2022-04-15 浙江力石科技股份有限公司 Big data tourism resource data compensation method
CN114091357A (en) * 2022-01-19 2022-02-25 杭州海康威视数字技术股份有限公司 Traffic flow prediction method, traffic flow prediction device, electronic device, and storage medium

Also Published As

Publication number Publication date
CN111179591B (en) 2021-02-26

Similar Documents

Publication Publication Date Title
CN111179591B (en) Road network traffic time sequence characteristic data quality diagnosis and restoration method
US11403164B2 (en) Method and device for determining a performance indicator value for predicting anomalies in a computing infrastructure from values of performance indicators
CN106991145B (en) Data monitoring method and device
Lindemann et al. Anomaly detection and prediction in discrete manufacturing based on cooperative LSTM networks
Manco et al. Fault detection and explanation through big data analysis on sensor streams
US11288577B2 (en) Deep long short term memory network for estimation of remaining useful life of the components
CN111045894B (en) Database abnormality detection method, database abnormality detection device, computer device and storage medium
CN116757534B (en) Intelligent refrigerator reliability analysis method based on neural training network
CN110865929A (en) Abnormity detection early warning method and system
CN112131212A (en) Hybrid cloud scene-oriented time sequence data anomaly prediction method based on ensemble learning technology
JP2017004509A (en) Advanced analytical infrastructure for machine learning
US10482204B2 (en) System for processing data and modelling for analysis of the energy consumption of a site
CN112036075A (en) Abnormal data judgment method based on environmental monitoring data association relation
CN109471698B (en) System and method for detecting abnormal behavior of virtual machine in cloud environment
US11675643B2 (en) Method and device for determining a technical incident risk value in a computing infrastructure from performance indicator values
Son et al. Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge
CN114978956A (en) Method and device for detecting abnormal performance mutation points of network equipment in smart city
CN105743595A (en) Fault early warning method and device for medium and short wave transmitter
CN113158448A (en) Self-adaptive calculation method for equipment threshold of ship system
CN110455370B (en) Flood-prevention drought-resisting remote measuring display system
RU2632124C1 (en) Method of predictive assessment of multi-stage process effectiveness
JP7062505B2 (en) Equipment management support system
CN116738192A (en) Digital twinning-based security data evaluation method and system
CN113688506B (en) Potential atmospheric pollution source identification method based on multi-dimensional data such as micro-station and the like
CN115856204A (en) Method for predicting gas concentration of tunneling working face based on three-dimensional echo state network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CP01 Change in the name or title of a patent holder
CP01 Change in the name or title of a patent holder

Address after: 310012 1st floor, building 1, 223 Yile Road, Hangzhou City, Zhejiang Province

Patentee after: Yinjiang Technology Co.,Ltd.

Address before: 310012 1st floor, building 1, 223 Yile Road, Hangzhou City, Zhejiang Province

Patentee before: ENJOYOR Co.,Ltd.